[FFmpeg-devel] [PATCH] libavcodec/hevc: add hevc idct_4x4_neon of aarch64

[xufuji456]

---
 libavcodec/aarch64/hevcdsp_idct_neon.S    | 51 +++++++++++++++++++++++
 libavcodec/aarch64/hevcdsp_init_aarch64.c |  4 ++
 2 files changed, 55 insertions(+)

diff --git a/libavcodec/aarch64/hevcdsp_idct_neon.S b/libavcodec/aarch64/hevcdsp_idct_neon.S
index 124c50998a..fe8baf1348 100644
--- a/libavcodec/aarch64/hevcdsp_idct_neon.S
+++ b/libavcodec/aarch64/hevcdsp_idct_neon.S
@@ -245,6 +245,54 @@ function hevc_add_residual_32x32_16_neon, export=0
         ret
 endfunc
 
+.macro tr_4x4 in0, in1, in2, in3, out0, out1, out2, out3, shift, tmp0, tmp1, tmp2, tmp3, tmp4
+         sshll           \tmp0, \in0, #6
+         smull          \tmp2, \in1, v4.h[1]
+         mov            \tmp1, \tmp0
+         smull          \tmp3, \in1, v4.h[3]
+         smlal          \tmp0, \in2, v4.h[0] //e0
+         smlsl          \tmp1, \in2, v4.h[0] //e1
+         smlal          \tmp2, \in3, v4.h[3] //o0
+         smlsl          \tmp3, \in3, v4.h[1] //o1
+
+         add            \tmp4, \tmp0, \tmp2
+         sub            \tmp0, \tmp0, \tmp2
+         add            \tmp2, \tmp1, \tmp3
+         sub            \tmp1, \tmp1, \tmp3
+         sqrshrn        \out0, \tmp4, #\shift
+         sqrshrn        \out3, \tmp0, #\shift
+         sqrshrn        \out1, \tmp2, #\shift
+         sqrshrn        \out2, \tmp1, #\shift
+.endm
+
+.macro transpose_4x4 r0, r1, r2, r3
+        trn1            v22.8h, \r0\().8h, \r1\().8h
+        trn2            v23.8h, \r0\().8h, \r1\().8h
+        trn1            v24.8h, \r2\().8h, \r3\().8h
+        trn2            v25.8h, \r2\().8h, \r3\().8h
+        trn1            \r0\().4s, v22.4s, v24.4s
+        trn2            \r2\().4s, v22.4s, v24.4s
+        trn1            \r1\().4s, v23.4s, v25.4s
+        trn2            \r3\().4s, v23.4s, v25.4s
+.endm
+
+.macro idct_4x4 bitdepth
+function ff_hevc_idct_4x4_\bitdepth\()_neon, export=1
+        ld1             {v0.4h-v3.4h}, [x0]
+
+        movrel          x1, trans
+        ld1             {v4.4h}, [x1]
+
+        tr_4x4          v0.4h, v1.4h, v2.4h, v3.4h, v16.4h, v17.4h, v18.4h, v19.4h, 7, v10.4s, v11.4s, v12.4s, v13.4s, v15.4s
+        transpose_4x4   v16, v17, v18, v19
+
+        tr_4x4          v16.4h, v17.4h, v18.4h, v19.4h, v0.4h, v1.4h, v2.4h, v3.4h, 20 - \bitdepth, v10.4s, v11.4s, v12.4s, v13.4s, v15.4s
+        transpose_4x4   v0, v1, v2, v3
+        st1             {v0.4h-v3.4h}, [x0]
+        ret
+endfunc
+.endm
+
 .macro sum_sub out, in, c, op, p
   .ifc \op, +
         smlal\p         \out, \in, \c
@@ -578,6 +626,9 @@ function ff_hevc_idct_16x16_\bitdepth\()_neon, export=1
 endfunc
 .endm
 
+idct_4x4 8
+idct_4x4 10
+
 idct_8x8 8
 idct_8x8 10
 
diff --git a/libavcodec/aarch64/hevcdsp_init_aarch64.c b/libavcodec/aarch64/hevcdsp_init_aarch64.c
index 88a797f393..1deefca0a2 100644
--- a/libavcodec/aarch64/hevcdsp_init_aarch64.c
+++ b/libavcodec/aarch64/hevcdsp_init_aarch64.c
@@ -49,6 +49,8 @@ void ff_hevc_add_residual_32x32_10_neon(uint8_t *_dst, const int16_t *coeffs,
                                         ptrdiff_t stride);
 void ff_hevc_add_residual_32x32_12_neon(uint8_t *_dst, const int16_t *coeffs,
                                         ptrdiff_t stride);
+void ff_hevc_idct_4x4_8_neon(int16_t *coeffs, int col_limit);
+void ff_hevc_idct_4x4_10_neon(int16_t *coeffs, int col_limit);
 void ff_hevc_idct_8x8_8_neon(int16_t *coeffs, int col_limit);
 void ff_hevc_idct_8x8_10_neon(int16_t *coeffs, int col_limit);
 void ff_hevc_idct_16x16_8_neon(int16_t *coeffs, int col_limit);
@@ -119,6 +121,7 @@ av_cold void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth)
         c->add_residual[1]             = ff_hevc_add_residual_8x8_8_neon;
         c->add_residual[2]             = ff_hevc_add_residual_16x16_8_neon;
         c->add_residual[3]             = ff_hevc_add_residual_32x32_8_neon;
+        c->idct[0]                     = ff_hevc_idct_4x4_8_neon;
         c->idct[1]                     = ff_hevc_idct_8x8_8_neon;
         c->idct[2]                     = ff_hevc_idct_16x16_8_neon;
         c->idct_dc[0]                  = ff_hevc_idct_4x4_dc_8_neon;
@@ -168,6 +171,7 @@ av_cold void ff_hevc_dsp_init_aarch64(HEVCDSPContext *c, const int bit_depth)
         c->add_residual[1]             = ff_hevc_add_residual_8x8_10_neon;
         c->add_residual[2]             = ff_hevc_add_residual_16x16_10_neon;
         c->add_residual[3]             = ff_hevc_add_residual_32x32_10_neon;
+        c->idct[0]                     = ff_hevc_idct_4x4_10_neon;
         c->idct[1]                     = ff_hevc_idct_8x8_10_neon;
         c->idct[2]                     = ff_hevc_idct_16x16_10_neon;
         c->idct_dc[0]                  = ff_hevc_idct_4x4_dc_10_neon;
-- 
2.32.0 (Apple Git-132)